Digital Biomarkers for Depression Screening With Wearable Devices: Cross-sectional Study With Machine Learning Modeling

• DOI: 10.2196/24872
• Publication Date: 2021-07-15T16:55:11Z
• Authors: Yuri Rykov; Thuan-Quoc Thach; Iva Bojic; George Christopoulos; Josip Car
• Keywords: Patient Health Questionnaire; Depression; Activity tracker; Cross-sectional study

Background Depression is a prevalent mental disorder that undiagnosed and untreated in half of all cases. Wearable activity trackers collect fine-grained sensor data characterizing the behavior physiology users (ie, digital biomarkers), which could be used for timely, unobtrusive, scalable depression screening. Objective The aim this study was to examine predictive ability biomarkers, based on from consumer-grade wearables, detect risk working population. Methods This cross-sectional 290 healthy adults. Participants wore Fitbit Charge 2 devices 14 consecutive days completed health survey, including screening depressive symptoms using 9-item Patient Health Questionnaire (PHQ-9), at baseline weeks later. We extracted range known novel biomarkers physical activity, sleep patterns, circadian rhythms wearables steps, heart rate, energy expenditure, data. Associations between severity were examined with Spearman correlation multiple regression analyses adjusted potential confounders, sociodemographic characteristics, alcohol consumption, smoking, self-rated health, subjective loneliness. Supervised machine learning statistically selected predict symptom status). varying cutoff scores an acceptable PHQ-9 score define group different subsamples classification, while set remained same. For performance evaluation, we k-fold cross-validation obtained accuracy measures holdout folds. Results A total 267 participants included analysis. mean age 33 (SD 8.6, 21-64) years. Out participants, there mild female bias displayed (n=170, 63.7%). majority Chinese (n=211, 79.0%), single (n=163, 61.0%), had university degree (n=238, 89.1%). found greater robustly associated variation nighttime rate AM 4 6 AM; it also lower regularity weekday steps estimated nonparametric interdaily stability autocorrelation as well fewer steps-based daily peaks. Despite several reliable associations, our evidence showed limited whole sample However, balanced contrasted comprised depressed no or minimal symptoms), model achieved 80%, sensitivity 82%, specificity 78% detecting subjects high depression. Conclusions Digital have been discovered are behavioral physiological consumer increased assist screening, yet current shows ability. Machine models combining these discriminate individuals risk.